Experiments & methodology

Once a hypothesis exists, the agent can turn it into a lab-ready experimental plan — a structured methodology where every design choice is grounded in the life-science literature, and anything the evidence does not cover is marked Not reported rather than invented.

#From evidence to a plan

A methodology is the bridge between a claim and a benchtop. The agent reads across the papers it has gathered on your question, extracts the concrete methodological parameters they report — sample sizes, model systems, assays, statistical approaches — and synthesises them into a coherent protocol designed to test a specific hypothesis. The output is not a generic template: each recommendation is tied back to the studies that justify it.

In practice this sits at the end of a longer chain. The agent first assembles a verified literature base, surfaces the mechanistic links inside it, and proposes a falsifiable hypothesis. The methodology is the step that makes that hypothesis testable, mapping cleanly onto the Research Plan section a grant reviewer or protocol committee expects to see.

#What a methodology covers

A generated methodology is organised into the components a peer reviewer looks for. Each block is a structured recommendation rather than a prose paragraph, so it can be lifted directly into a protocol or grant methods section and adapted.

#Design and model-system justification

The study design — type, arms, timeline, and any randomisation or blinding strategy, with a preregistration recommendation where appropriate. Alongside it, a justified model system: the recommended cell or animal model with species, sex, age, and passage detail, plus the scientific rationale for choosing it over alternatives.

#Sample size and statistical power

An explicit power analysis: alpha level, target statistical power, effect-size estimates, and replicate numbers — anchored, wherever possible, to the values reported in comparable published studies rather than assumed.

#Interventions, assays, controls and replicates

The experimental perturbations and the key assays used to read them out, with brand or catalogue names where the literature specifies them. Every intervention is paired with its controls — positive, negative, and sham as relevant — and a recommendation for technical and biological replicates.

#Endpoints with Go/No-Go criteria

Primary metrics with quantitative thresholds, secondary endpoints, and explicit success and futility rules. Framing endpoints as Go/No-Go decisions makes the plan actionable: it states in advance what result would justify advancing the work and what result would halt it.

#Statistical analysis and confounders

A statistical analysis plan — recommended tests and their assumptions, effect sizes, confidence intervals, and corrections for multiplicity — together with the confounding factors the agent identifies from the evidence and concrete strategies to mitigate each one.

#Grounding and the "Not reported" integrity rule

Every element of a methodology carries a citation marker that tells you where it came from and how directly. The agent distinguishes two kinds of support:

• Direct evidence — a source paper explicitly reports the methodological approach being recommended.
• Derived reasoning — the recommendation is synthesised from methodological details across one or more cited papers rather than lifted from a single one.

Each component therefore ends with a marker such as (Direct; PMID:12345678) or (Derived; PMID:12345678, PMID:87654321), so any parameter can be traced to the evidence that justifies it.

The same discipline governs citations. References are grounded in the life-science literature — 40M+ papers spanning published journals and preprints — and checked at generation time. Where the proposed combination of reagents, model, and assay is genuinely novel, the agent flags the relevant citations as unverified rather than dressing the plan in borrowed authority. In a methodology, that is often the point: a novel protocol is expected to extend beyond what any single prior paper has done.

#Fundamental vs applied designs

The agent shapes a methodology to the kind of research you are doing. The same hypothesis can demand a very different plan depending on whether the work is basic-science or translational, and the agent orients its recommendations accordingly.